Machine learning assisted classification of post-treatment amines for increasing the stability of organic-inorganic hybrid perovskites

The poor environmental stability of hybrid organic-inorganic perovskites (HOIPs) is the main reason hindering its application. The post-treatment of HOIPs with amines has gradually become a simple and effective protection strategy. Here, the support vector classifier (SVC) was proposed to select post-treatment amines for increasing the stability of HOIPs. Based on the seven optimal features, the accuracies of leave-one-out cross-validation (LOOCV) and test set of SVC are 95.45% and 90.91%, respectively. Additionally, the Shapley additive explanation (SHAP) was introduced to explain the model, and the results indicated that the features nHDon, PW2, Mv and MLOGP play the important roles in the prediction of amine molecular classes. Through the analysis of models, we found that amines with fewer hydrogen-bond donor atoms, larger mean atomic van der Waals volumes (scaled on Carbon atom), more hydrophobic groups, more branched chains, and smaller van der Waals surface area of atoms can be successfully used as potential candidates. Moreover, we successfully screened out 13718 candidate amines from tens of thousands of amines, which are expected to be the preferred post-treatment amines for HOIPs.